Research On The Looseness Fault Characteristic And Recognition Method

This paper carries out the research work on the cable joint looseness fault in the low voltage power distribution line.Study the typical characteristics of the cable joint looseness faults deeply,and propose the fault recognition methods.This is very great practical significance for the reliable operation of the power supply and distribution system.In order to further research on the typical features of the cable joint looseness and realize fault identification,lots of looseness fault experiments of copper-copper electric connector under different current and loosing states are carried out by self-developed experimental platform.The temperature characteristics,contact voltage and current characteristics of the loosening bolted cable joint under different conditions are studied.The validity of the two kind methods for detecting the looseness faults is compared.One is based on wavelet energy entropy recognition method.The multi-resolution analysis of current signal is conducted by using wavelet transform and the current energy entropy used as typical feature parameters of electrical connection looseness fault are extracted.The second is based on the bilinear time frequency analysis method.First,using the Wigner-Ville distribution to transform the contact current and acquire the combined time-frequency characteristics.Then extract the time edge and the frequency edge of the signal.Then calculate the loop current's average energy and fundamental frequency energy.The distribution of the Hough transform is carried out,namely complete Wigner-Hough transform(WHT),and the peak and peak change rate interval length are obtained.Using the average energy,fundamental frequency energy and WHT peak as the characteristic parameters to carry out the fault identification.The input vectors of the model are obtained by using the characteristic parameters of these two methods respectively.The fault diagnosis is carried out using the optimized probabilistic neural network(PNN)identification model.At last analyzes the influence of parameter S and the sample size on the recognition result.The results show that: the accuracy of wavelet energy entropy algorithm is low,but the real-time is higher;bilinear time-frequency algorithms of higher precision,but lower real-time performance,therefore according to the actual requirement to choose the different method.